Theorem 9 — any (collective) scheme learning passive states with needs
ProvedBosonicLearn.theorem_9bosonic-gaussian-statesp2o-gran-per-chapterp2o-plan-paperp2o-v1quantum-informationquantum-learning
There are constants and such that for every , every and every : any scheme, including arbitrary collective (entangled) measurements on , that learns every -mode passive Gaussian state with to trace distance with probability needs
It is the warm-state half of the lower bound in Theorem 3: copies are necessary even for entangled measurements.
Differences from the source:
- reading: "for any ν > 0, 0 < ε < ε₀, and sufficiently large n" → n₀ chosen before ν and ε; reason: the stronger reading; the proof's "sufficiently large n" (ensemble (223), Fano (229)) does not depend on ν or ε, and Theorem 3 needs it uniform
- encoding: "any scheme", "arbitrary collective measurements" → a POVM on the N-copy Fock space whose outcome is the estimated density operator; reason: measurement followed by any classical post-processing is again such a POVM
Preamble
import Mathlib import Definitions.Def_BosonicLearn_collectiveSuccess import Definitions.Def_BosonicLearn_promiseClass import Definitions.Def_BosonicLearn_IsPassive open QAlgorithms.Papers.Independent.Dependency50.P33 (State traceSize) open QAlgorithms.Papers.Independent.Claimed50.P47 (Occupation) open MeasureTheory ProbabilityTheory open scoped ENNReal Matrix
Formal statement
namespace BosonicLearn
/-- **Theorem 9 (Matching lower bound for warm states, any collective scheme).**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, Theorem 9, p. 65: "There exist constants C > 0 and ε_0 > 0 such that the following holds: For any ν > 0,
0 < ε < ε_0, and sufficiently large n, any scheme that can learn any passive Gaussian state ρ(0, Σ) to ε trace
distance with 2/3 probability with the promise that Σ ≥ (1/2 + ν)I requires at least N ≥ Cn²/ε² copies."
The proof covers "arbitrary collective measurements" (p. 67).
There are `C, ε₀ > 0` and `n₀` such that for every `ν > 0`, `ε ∈ (0, ε₀)`, `n ≥ n₀`, and every POVM on `N` copies
(entangled across copies) whose outcome is the estimate: if every passive Gaussian state with `Σ ⪰ (½ + ν)I` is
learned to trace distance `ε` with probability at least `2/3`, then `N ≥ Cn²/ε²`.
**Formalization Note.** "Sufficiently large n" is read uniformly in `ν` and `ε` (`n₀` before `ν, ε`); see the
Differences list. -/
theorem theorem_9 :
∃ C : ℝ, 0 < C ∧ ∃ ε₀ : ℝ, 0 < ε₀ ∧ ∃ n₀ : ℕ, ∀ ν : ℝ, 0 < ν → ∀ ε : ℝ, 0 < ε → ε < ε₀ →
∀ n : ℕ, n₀ ≤ n →
∀ (N : ℕ) (E : FockPOVM (Fin N → Occupation (Fin n)) (State (Occupation (Fin n)))),
(∀ ρ ∈ promiseClass n ν, IsPassive ρ → ENNReal.ofReal (2 / 3) ≤ collectiveSuccess E ρ ε) →
C * ((n : ℝ) ^ 2 / ε ^ 2) ≤ N := by sorry
end BosonicLearn
Source
Chen, Anna Mele, Anna Mele, Preskill, When are bosonic Gaussian states classical to learn?, arXiv:2609.26705, Theorem 9, p. 65 (collective measurements, p. 67)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.